Skip to main content
OpenAIProvider wraps the official OpenAI Python SDK. It supports non-streaming and streaming chat, tool calling, and multimodal (vision) prompts for compatible models.

​
Installation

1

Install Python dependencies

pip install logicore openai
2

Set your API key

​
Constructor parameters

from logicore.providers.openai_provider import OpenAIProvider

provider = OpenAIProvider(model_name="gpt-4o-mini")
​
model_name
string
required
The OpenAI model ID to use. Examples: "gpt-4o", "gpt-4o-mini", "gpt-4-turbo", "o1-mini". Must be a model available in your OpenAI project.
​
api_key
string
Your OpenAI secret key. If omitted, the provider reads OPENAI_API_KEY from the environment. Raises ValueError if neither is set.
​
**kwargs
any
Extra keyword arguments forwarded to the underlying openai.OpenAI() client constructor. Use this to set a custom base_url (e.g., for OpenAI-compatible endpoints), timeout, max_retries, or http_client.

​
Basic usage

import asyncio
from logicore.agents.agent import Agent
from logicore.providers.openai_provider import OpenAIProvider

async def main():
    provider = OpenAIProvider(model_name="gpt-4o-mini")

    agent = Agent(
        llm=provider,
        role="Cloud Assistant",
        system_message="Provide concise and reliable answers."
    )

    result = await agent.chat("Explain event-driven architecture in simple terms.")
    print(result)

asyncio.run(main())

​
Streaming

Provide an on_token callback to receive each token as it arrives. The callback can be synchronous or async: 
import asyncio
from logicore.providers.openai_provider import OpenAIProvider

async def main():
    provider = OpenAIProvider(model_name="gpt-4o-mini")

    collected = []

    def on_token(token: str):
        print(token, end="", flush=True)
        collected.append(token)

    result = await provider.chat_stream(
        messages=[
            {"role": "user", "content": "List 5 principles of clean code."}
        ],
        on_token=on_token
    )

    print()  # newline
    # result is a ChatCompletionMessage with .content and .tool_calls
    print("Tool calls:", result.tool_calls)

asyncio.run(main())
chat_stream assembles the full response from streamed chunks and returns a ChatCompletionMessage object once streaming completes. Tool-call chunks are reconstructed correctly.

​
Tool calling

def calculate_tax(amount: float, rate: float) -> str:
    """Calculate tax amount from value and rate."""
    return str(amount * rate)

agent = Agent(
    llm=OpenAIProvider(model_name="gpt-4o-mini"),
    tools=[calculate_tax]
)

result = await agent.chat("What is the tax on $200 at 8.5%?")
print(result)
When tools are provided, the provider sets tool_choice="auto" so the model decides when to call a tool.

​
Vision / multimodal

Use a vision-capable model (gpt-4o, gpt-4o-mini) and pass a list of content parts: 
import asyncio
from logicore.agents.agent import Agent
from logicore.providers.openai_provider import OpenAIProvider

async def main():
    agent = Agent(
        llm=OpenAIProvider(model_name="gpt-4o-mini"),
        role="Vision Assistant"
    )

    message = [
        {"type": "text", "text": "What is shown in this image?"},
        {"type": "image_url", "image_url": "https://upload.wikimedia.org/wikipedia/commons/thumb/4/47/PNG_transparency_demonstration_1.png/280px-PNG_transparency_demonstration_1.png"}
    ]

    result = await agent.chat(message)
    print(result)

asyncio.run(main())
Supported image_url values:
  • Local file path
  • https:// image URL
  • data:image/...;base64,... inline data

​
Using OpenAI-compatible endpoints

Pass a custom base_url to target any OpenAI-compatible API (e.g., LM Studio, Together AI, Anyscale): 
provider = OpenAIProvider(
    model_name="mistralai/Mixtral-8x7B-Instruct-v0.1",
    api_key="your-key",
    base_url="https://api.together.xyz/v1"
)

​
Troubleshooting

The api_key argument was not passed and OPENAI_API_KEY is not set in the environment. Set the environment variable or pass api_key directly to the constructor.
The model returned a response with neither text content nor tool calls. This can happen with certain system prompts or with reasoning models in restricted modes. Check your system_message and ensure the model you selected is available in your project.
Double-check the key starts with sk- and matches a live key in your OpenAI dashboard. Keys are not shared across organisations.
You have hit your usage tier’s rate or spend limit. Either add credits, request a tier upgrade, or implement retries with exponential back-off using the max_retries kwarg on the client.
Some models (e.g. gpt-4) require explicit access in the OpenAI console. Use gpt-4o-mini for broad availability or verify the model is enabled for your project.

Build docs developers (and LLMs) love

Get started for freeTalk to us